No single setting stops cheating on an online assessment. Employers that trust their results combine three layers: test design that makes shared answers useless, monitoring that records what happened during the test, and a later step that checks whether the candidate can do the work in person. Each layer covers gaps in the others, and the last one decides whether the score is believable. That combination is what an AI Candidate Evaluation process is built around.
1. Design the test so answers don't travel
- Randomize questions. Draw each candidate's questions from a larger pool, or at least shuffle the order and the multiple-choice options. Vervoe, Testlify and Coderbyte all support randomization, so a leaked answer key matches few candidates.
- Set time limits. A timed test leaves less room to look answers up. Some test providers sell timed and untimed versions of the same test; the timed one gives the shorter window.
- Use work samples. Job-simulation tasks that ask candidates to apply their own experience are harder to copy than fact questions.
2. Monitor the session
Most assessment platforms offer some mix of these, configured per test:
- Copy-paste tracking or blocking, so large blocks of text or code appearing at once get flagged.
- Tab-switch and focus tracking, which logs when a candidate leaves the test window.
- Webcam snapshots or video, to confirm the same person sat the whole test.
- Screen recording for higher-stakes roles.
- Plagiarism and similarity checks for coding tests, comparing each solution with other candidates' submissions.
Treat flags as prompts for review, not verdicts. A single tab switch can be a notification; repeated exits plus an unusually fast score is a pattern worth a closer look. Tell candidates up front which monitoring is on. It deters casual cheating and avoids complaints later.
3. Verify the result later
The strongest control happens after the test. A short live interview that asks candidates to walk through their answers, or to extend a coding solution, shows quickly whether they understand the work they submitted. Some coding platforms now add AI-generated follow-up questions about each candidate's own submission for the same reason. A candidate who passed with outside help usually struggles to explain their reasoning.
A practical setup
For most hiring teams:
- Randomize every test that runs at volume.
- Keep copy-paste and tab tracking on by default; add webcam or screen recording for roles where a wrong hire is expensive.
- Review flagged sessions before rejecting anyone.
- Add one live verification step for every candidate you plan to hire.
That combination keeps the test fair for honest candidates and makes the score something you can rely on.